Abstract: Addressing the issue of band redundancy in multispectral/hyperspectral remote sensing images for power line inspection, this paper proposes a band selection-based image classification method. A sample set is constructed through data augmentation, and three methods—Maximum Variance principal component analysis, Improved sparse subspace clustering, and density peaks clustering—are used to extract discriminative band features. These features are then evaluated using Support Vector Machine(SVM) classification experiments. Results show that the ISSC method consistently outperforms others under 5, 10, and 15 band selection scenarios, significantly improving classification accuracy and demonstrating its advantage in band selection tasks.